GradTOD - A Unified Dialogue State Tracking Model for Task-Oriented and Open Domain Dialogues

The task-oriented dialogue domain system requires classifying intent and replying to a specific goal domain. In the sub-module of Task-oriented, the Dialogue State Tracker (DST) is well-known as a variety processing tracker. However, existing DST models often specialize in only task-oriented domains (ToD), leading to limited performance when applied to scenarios. In this paper, we propose GradTOD, a unified DST model that predicts both two task types, task-oriented dialogue (TOD) and open-domain dialogue (ODD). Our model leverages the recent advances in prompt engineering and conditional generation to perform zero-shot learning. After experiments, GradTOD has achieved an 88.6% and 82.5% score on Joint Goal Accuracy metrics when evaluating the Scheme-Guided Dialogue (SGD) and FusedChat test sets correspondingly, demonstrating the adaption ability for multi-domains.